Latest / Mechanical Engineering Made Simple / (#124) Why Your Precision Parts Don't Fit

Transcript

(#124) Why Your Precision Parts Don't Fit · Mechanical Engineering Made Simple

0:00
Welcome back design junkies to Mechanical Engineering Made
0:02
Simple. If the podcast brings you value
0:04
help, support the show with a like a follow or leave a comment
0:07
below. Thanks for having me back.
0:09
So I want you to imagine the scenario.
0:11
You are standing in a massive hyper clean assembly Bay, right?
0:15
OK, setting the scene. Yeah.
0:17
And above you is a payload fairing.
0:20
It's preparing to close over this state-of-the-art
0:23
communications satellite. I mean, this is a massive
0:25
aerospace project where the launch logistics alone cost
0:29
10,000 a pound. Wow, right?
0:31
Every single gram has been optimized, every subsystem has
0:35
been meticulously modeled, and as the technicians move in to,
0:39
you know, secure that final maiden bracket between the solar
0:43
array deployment mechanism and the main chassis I.
0:46
Guess the bolts don't align. Exactly.
0:48
The holes are off by a fraction of a millimeter.
0:50
The pin just will not seat and you can't just grab a hand drill
0:54
and bore out the hole on a flight ready spacecraft.
0:57
Oh, absolutely not. You'd be fired on the spot.
0:59
Right, so you are entirely dead in the water.
1:02
You're bleeding hundreds of thousands of dollars a day in
1:05
delays, all because two pieces of metal, manufactured on
1:08
opposite sides of the country, mind you, simply refuse to fit
1:12
together. Yeah, I mean, that is the
1:14
ultimate nightmare scenario for literally any project manager or
1:19
design engineer. And unfortunately, it happens
1:22
far more often than the industry really likes to admit.
1:25
Oh, I bet. And that's exactly what we're
1:26
looking at today. We've got a really interesting
1:28
stack of sources that address this exact pain point.
1:32
Yeah, we've got some heavy hitters today.
1:33
We do. We're looking at a highly
1:35
detailed tolerance analysis report from Boeing Research and
1:38
Technology authored by Fritz Schultz.
1:41
That one really digs deep into the statistical mathematics of
1:44
how parts stack together. The heavy math.
1:46
Exactly. Plus we've got some very
1:48
practical sourcing and manufacturing guidelines from
1:51
zometry, and we're also taking a look into the historical
1:53
evolution of NASA and ASME engineering drawing standards.
1:57
And the overarching theme across all these documents is just the
2:01
brutal physical reality of manufacturing variation and, you
2:06
know, the mathematical language we have to develop to control.
2:08
It right. And that language is geometric
2:10
dimensioning and tolerancing, universally known as GD and T
2:14
That's the one, along with what I like to call the dark art of
2:18
tolerance stack analysis. Because whether you are a
2:21
seasoned mechanical engineer who spends, you know, 8 hours a day
2:25
detailing CAD models, or you're someone adjacent to
2:29
manufacturing like in quality control, procurement, or even
2:32
project management. Right, Anyone touching the
2:34
hardware? Exactly understanding the why
2:37
and how of this topic is basically your ultimate
2:39
shortcut. It is the shortcut to avoiding
2:42
that exact Houston we have a problem moment on the assembly
2:46
floor. Precisely because when you are
2:47
dealing with mass manufacturing, or like you said, high stakes
2:51
aerospace where you're paying £10,000 for orbit insertion,
2:54
perfection is just a physical impossibility.
2:57
Right, you literally can't make a perfect part.
2:59
You can't. The machine tools wear down, the
3:02
temperature of the cutting fluid fluctuates during the day.
3:05
The metal itself literally springs back a little bit after
3:09
it's been machined. Oh right, internal stresses.
3:12
Exactly. So you just cannot demand
3:15
perfect parts. What you must demand, and what
3:17
GD and T actually allows you to achieve, is parts that are
3:21
perfectly understood within a really highly specific framework
3:24
of allowed imperfections. Allowed imperfections.
3:27
I like that. Yeah, and to truly understand
3:29
why this framework is so incredibly rigid today, we
3:33
actually need to strip away the modern CAD software for a minute
3:37
and look at how this problem almost broke the industrial
3:40
supply chain completely. Oh I'm glad you brought up the
3:42
history because there is this massive common misconception I
3:46
hear especially from junior designers.
3:48
Oh, what's that? Well, they think GD and T is
3:50
just like an academic exercise, like it's something a group of
3:54
grumpy mathematicians dreamed up just to make reading blueprints
3:57
harder for everyone else. Right, like it's just busy work.
3:59
Exactly. But it was actually born out of
4:01
sheer panic, wasn't it? Absolute panic.
4:03
I mean, if you rewind to the mid 20th century specifically during
4:07
that massive scale up of military and industrial
4:10
production surrounding the world wars, the whole nature of
4:14
manufacturing fundamentally changed.
4:16
How so? Well before that, a lot of
4:19
complex machinery was built under one roof.
4:22
The machine is making the piston was literally in the same
4:25
building as the machine is making the engine block.
4:28
So they could just talk to each other.
4:29
Exactly. If a piston was a little tight,
4:31
they just walked across the shop floor and figured it out with
4:33
some sandpaper or a file and fitting.
4:36
But mass production completely decentralized that supply chain.
4:39
Suddenly, you know, the Pistons are being cast in Ohio, they're
4:42
being machined in Michigan, and the engine blocks are coming
4:45
from Pennsylvania. So they were completely relying
4:48
on traditional engineering drawings to communicate between
4:51
these states, Yes. What we now call CD and T or
4:54
coordinate dimensioning and tolerancing, right?
4:56
The old school basic dimensions. And CD and T proved to be
5:00
catastrophically inadequate. The defective parts problem
5:03
became astronomical. Wow, really?
5:06
Oh yeah, the military and the industrial sector were receiving
5:09
thousands of parts that perfectly matched the numbers
5:12
written on the drawing but completely failed to assemble in
5:15
the field. Wait, how is that possible?
5:17
If the numbers matched, why didn't they fit?
5:19
Because they realized that a basic 2D drawing with standard
5:23
linear tolerances simply does not reflect the actual
5:27
functional relationship between mating parts.
5:29
OK, wait. Let's dig into why that
5:31
traditional CD and T drawing fails so spectacularly, because
5:34
I feel like this is where the illusion of the blueprint really
5:38
falls apart for people. It really does.
5:39
So let's say I'm a junior engineer, right?
5:42
I open up my CAD software and I draw a simple rectangular block
5:46
and I put a hole in it. I use the basic dimension tool
5:49
to say the hole is exactly 2 inches from the left edge and I
5:52
type in a tolerance of ±0.05 inches.
5:56
A classic CD and T call. Out right to my mind, I have
6:00
done my job perfectly. I gave the machinist a target
6:02
and I gave them a mathematical boundary.
6:04
Why is that an illusion? It's an illusion because your
6:07
CAD software exists in a perfect theoretical universe, and that
6:12
physical piece of aluminum sitting on the milling machine
6:15
simply does not. OK, Fairpoint.
6:18
The misconception there is that those standard linear tolerance
6:22
zones, you know that little plus or minus box you typed in
6:26
actually define the real world geometry of the part.
6:29
Wait, they don't. They do not.
6:31
Traditional CD and T only gives you a linear distance between
6:34
two theoretical points. It completely ignores the actual
6:38
form, the overall shape, and the physical orientation of the
6:42
feature. Wait, break that down for me.
6:43
If I take my tallipers and I measured 2 inches from the edge
6:46
and it says 2 inches, isn't that the whole story?
6:48
Not even close. Let's look at your rectangular
6:50
block. Yeah, you measured 2 inches from
6:51
the left edge, right? Yeah.
6:53
But is that physical left edge perfectly straight?
6:55
In reality it's going to have microscopic hills and valleys
6:58
from the end mill that cut it. Oh sure, Surface finish.
7:01
Right. It might be slightly bowed, it
7:03
might not be perfectly perpendicular to the bottom
7:05
surface. O.
7:06
When the quality inspector places their measurement tool
7:09
against that edge, which Microscoic Hill are they
7:13
referencing? Oh, I see.
7:14
Are they measuring from the top corner, the bottom corner, or
7:17
the middle of the bow? Your simple plus or minus
7:20
tolerance provides absolutely no clear frame of reference for how
7:24
the part actually functions in an assembly.
7:26
Or how it has to be physically oriented to be inspected.
7:29
That makes total sense. Exactly.
7:31
Actually, the source material has a brilliant concrete example
7:34
of this breaking down. They show a drawing of a flanged
7:37
cylindrical part. Oh yeah, the top hat example.
7:40
Right. Imagine a top hat.
7:41
You have a central hole going all the way through it, a raised
7:43
hub in the middle, and a wider outside diameter flange at the
7:47
bottom. Got it.
7:48
And the drawing calls out this small slot cut into the outer
7:52
edge of the flange. It gives a dimension for the
7:54
location of that slot with a tolerance of ±0.4mm.
7:59
Seems simple enough. Right, it looks completely
8:01
unambiguous on paper. But on the factory floor it is
8:05
an absolute nightmare. The question immediately
8:08
becomes, what is the origin of that dimension?
8:13
Where is the 00 coordinate exactly?
8:16
Is that plus or minus OH 4mm measured from the center axis of
8:21
the inner bore? Is it measured from the center
8:23
axis of the raised hub or is it from the center axis of the
8:27
outside diameter flange? And see, in my perfect CAD
8:31
model, all three of those axes are perfectly concentric.
8:35
They share the exact same mathematical line in digital
8:37
space. They do.
8:39
So as the designer. I don't even think to specify
8:41
which one it could because to me, looking at the screen, they
8:43
are all the exact same thing. Precisely.
8:46
But when a machinist chucks that piece of raw steel into a lathe,
8:50
turns the outside diameter, then physically flips the part around
8:54
to bore out the inner hole, tiny amounts of run out and variation
8:57
are introduced. Oh because they re clamped it.
8:59
Exactly. The inner hole, the raised hub,
9:01
and the outside flange will all be slightly eccentric to one
9:04
another. Their physical center axis will
9:06
be scattered by fractions of a millimeter, so when the quality
9:09
inspector tries to verify the location of your slot, their
9:12
measurement depends entirely on which physical feature they use
9:16
to establish their center point. So if they use the inner bore as
9:20
zero, the slot might pass inspection.
9:22
Yep. And if they use the outer
9:24
flange, it might fail and the CD and T drawing doesn't tell them
9:29
which one is the functional requirement for the assembly.
9:31
And this vacuum of information is exactly what breeds what the
9:35
sources call tribal knowledge. Oh, tribal knowledge.
9:38
Yeah, the enemy of mass production.
9:40
Because the drawing is fundamentally ambiguous, the
9:43
quality inspector literally has to stop what they're doing.
9:46
Walk across the facility to the engineering department, track
9:50
down the original designer and ask, hey, when you drew this
9:53
slot, but what did you actually intend for it to line up with?
9:56
Which is just a complete system failure.
9:58
An engineering drawing or a modern 3D model data set is
10:03
supposed to be a legally binding contract between the customer
10:06
and the supplier. If that contract is based on
10:08
tribal knowledge, like on one guy remembering a conversation
10:11
he had with another guy three years ago over coffee, you have
10:16
no legal recourse when the parts show up and do not fit.
10:19
Because under those ambiguous CD and T rules, a supplier can
10:23
manufacture a part that technically meets their valid
10:26
interpretation of your vague drawing.
10:28
Even if it's completely useless in your final assembly.
10:31
Exactly, and because they technically met the dimensional
10:34
numbers on the page, you the customer have to foot the bill
10:38
and pay for the scrap. Yep.
10:40
It makes me think of an analogy. It's kind of like trying to
10:42
scale up a family recipe. Oh, I like this.
10:44
Go on. Say my grandmother writes down.
10:46
Add a pinch of salt and bake until the crust is golden brown.
10:50
That works perfectly in her kitchen because she possesses
10:53
the tribal knowledge. Right, she knows her own
10:54
kitchen. She knows the exact humidity,
10:56
she knows her specific oven runs hot on the left side, and she
11:00
knows exactly what golden brown means to her eyes.
11:03
But if I take that exact piece of notebook paper, hand it to a
11:06
massive automated commercial food factory and ask them to
11:10
produce 10,000 pies a day. It's a total disaster.
11:14
Right one factory shift managers pinch is 20 grams, another's is
11:18
5g and golden brown under harsh LED factory lights looks
11:23
completely different than in a sunny home kitchen.
11:25
That is exactly the dynamic GD, and T is the engineering
11:29
equivalent of eliminating the pinch and the golden brown.
11:33
It replaces vague visual instructions with mathematically
11:37
undeniable parameters. So instead of bake until done.
11:40
It states bake at a sustained ambient temperature of 350°F for
11:45
exactly 42 minutes, verified by an internal thermal probe
11:49
calibrated to NIST standards. That is perfectly clear.
11:52
No tribal knowledge needed. Exactly.
11:54
GD and T uses a standardized vocabulary, highly specific
11:58
rules, and mathematically defined symbols to create a
12:01
legally indisputable definition of the physical part.
12:03
OK, but wait, if GD and T is this perfect, indisputable,
12:07
internationally recognized standard, why are there still so
12:10
many arguments on the shop floor?
12:11
Because reading through the sources, specifically looking at
12:14
how different regions apply these rules, it seems like we
12:16
just traded one set of problems for another.
12:19
You're not entirely wrong. Like we agreed to stop using the
12:22
grandmother's recipe, but now half the factory is speaking
12:24
French and the other half is speaking German and they don't
12:26
even realize it until the parts crash into each other.
12:29
You are hitting on one of the most insidious and frankly
12:32
expensive pain oints in modern global manufacturing.
12:36
The language. Barrier.
12:38
The global language barrier. We have two rimary global
12:41
standards that look incredibly similar on the surface.
12:44
They literally use the same visual symbols, but they operate
12:47
on fundamentally different mathematical philosophies deep
12:50
down. And we are talking about the
12:52
massive divergent between ASMEY 14.5 and ISO 11/01, right?
12:57
Exactly ASMEY 14.5 is heavily utilized in the United States
13:02
and by many massive multinational aerospace and
13:04
automotive corporations. Well, ISO eleven O 1 is the
13:07
dominant standard across Europe and parts of Asia.
13:10
But how bad could the divergent really be?
13:12
I mean, a circle is a circle, right?
13:14
You would think so. The sources highlight a specific
13:17
timeline that maps out what we can call the 65% problem, and
13:20
the numbers are frankly terrifying if you manage a
13:22
global supply chain. The historical drift between
13:26
these two organizations is startling.
13:29
If we look back to 1994, one of the great pioneers of this
13:32
field, Lowell Foster, published A definitive textbook called
13:36
Geometrics Thur. At that point in time, Foster
13:40
evaluated the two systems and claimed there was roughly a 90
13:44
to 95% agreement between the ASME standard and the ISO
13:48
standard. So they were basically
13:49
identical. They were on a path toward
13:51
global harmonization. A mechanical engineer in Detroit
13:54
could send a drawing to a CNC machinist in Munich and they
13:58
would both interpret the symbol. Almost identically.
14:00
But then we Fast forward. The sources bring up Alex
14:02
Krulikowski's 2010 book ISO Geometrical Tolerancing and The
14:06
Entire Landscape Has Fractured. Fractured is the right word.
14:09
Krulikowski's analysis states that roughly 65% of all possible
14:13
tolerance and specifications are now either specified differently
14:16
or interpreted differently between ASME and ISO.
14:20
Wait, really? We went from 5% disagreement to
14:23
65% disagreement. How does a global technical
14:26
language devolve that quickly? It wasn't really a devolution,
14:30
it was a philosophical divergent in response to rapidly advancing
14:34
technology. What do you mean?
14:35
Well, as manufacturing capabilities and coordinate
14:38
measuring machines became exponentially more precise, the
14:42
two standards committees had to decide how to mathematically
14:45
define increasingly complex geometry.
14:48
OK, they kept using the same visual symbols on the drawings
14:52
circle with a crosshair for osition 2, concentric circles
14:55
for concentricity, but the underlying software algorithms
14:59
they wrote to verify those symbols took completely
15:01
different paths. I need to understand the why
15:04
here. Why did they take different
15:05
paths? Is it just stubbornness or is
15:07
there a functional reason behind it?
15:09
It comes down to what each standard prioritizes.
15:12
Fundamentally, ASMEY 14.5 is heavily influenced by the
15:16
concept of the perfect functional boundary.
15:19
Functional boundary OK. It is deeply rooted in the idea
15:22
of assembly. It asks if I have a pin and a
15:25
hole, what is the absolute worst case envelope of that pin and
15:28
will it physically slide into the hole?
15:30
It thinks in terms of 3D mating envelopes.
15:32
Well, it's very practical. Will the parts go together?
15:35
Exactly ISO 11 O one, on the other hand, evolved with a
15:39
strong philosophical focus on independent geometric elements.
15:44
Meaning what exactly? It tends to look at the isolated
15:47
surface or the derived theoretical axis of a feature,
15:51
separate from how it interacts with the rest of the part.
15:53
It's more purely mathematical approach to the geometry itself.
15:57
OK, let's look at how that philosophical difference
15:59
actually ruins a part on the factory floor.
16:01
Give me a concrete example. Let's take the concentricity
16:04
symbol. It's two circles inside each
16:06
other. Right, so let's say I'm
16:07
designing a high speed spinning shaft.
16:09
It has a thick middle section and a thinner end section, and I
16:13
need them to share the exact same center point so the shaft
16:16
doesn't vibrate itself to pieces at 10,000 RPM.
16:19
OK, perfect example. I dropped the concentricity
16:21
symbol on the CAD drawing. How does an ISO inspector
16:24
evaluate that versus an ASME inspector?
16:27
If the drawing is governed by the ISO 1101 standard, that
16:31
concentricity symbol strictly commands the inspector to
16:34
evaluate the derived central axis of the feature.
16:37
So they just find the middle line.
16:38
Basically, they will probe the surface of the thinner section
16:41
of the shaft, use the software to calculate the theoretical
16:44
central line running down the middle of it, and then check if
16:46
that invisible line falls within the allowed tolerance zone
16:50
relative to the thicker section. ISO primarily cares about the
16:53
theoretical spine of the part. OK, that makes perfect sense
16:56
mathematically. So what does ASME do?
16:59
Under ASME why 14.5, that exact same visual symbol requires an
17:04
entirely different physical measurement process.
17:07
ASME concentricity controls the opposed median points of the
17:11
feature. Opposed median points.
17:13
Yes. To inspect it properly, the
17:15
quality technician cannot just calculate a central axis.
17:18
They have to map out thousands of distinct points exactly
17:22
opposite each other on the circumference of the shaft.
17:24
That sounds intense. It is.
17:26
They take point A on the top and point B exactly 180° across from
17:30
it on the bottom, and they find the exact midpoint between them.
17:33
They have to do this all the way around and all the way down the
17:35
shaft. They create a cloud of median
17:37
points, and every single one of those microscopic points must
17:40
fall within the tolerance zone. That sounds incredibly tedious.
17:44
Why would ASA me require that? Because of rotational mass, if a
17:47
shaft is slightly lobed, say it is shaped a tiny bit like a
17:51
triangle instead of a perfect circle, its theoretical central
17:55
axis might still be perfectly straight exactly ISO would pass
17:59
it. But if you spin that triangular
18:01
shaft at high speed, the unbalanced mass on the outside
18:05
will cause massive vibration. Right, because the weight isn't
18:08
distributed evenly even if the center line is straight.
18:11
Precisely by forcing you to measure the opposed median
18:14
points, ASME concentricity inherently controls both the
18:18
center location and the balance of the mass distribution.
18:22
It is a much stricter, more functional control for high
18:25
speed dynamic parts. That's a huge difference.
18:27
So if your supplier uses an ISO CMM program to measure an ASME
18:31
concentricity call out, they will just check the axis past
18:34
the load shaft. And your assembly will tear
18:36
itself apart on the testing. Man, that is terrifying.
18:38
And this brings up another critical divergent from the
18:40
sources, specifically regarding how we just measure basic size,
18:44
just the simple diameter of A. Yes, rule #1.
18:47
Right in ASME, why 14.5? A size tolerance inherently
18:52
controls 2 completely different things at the same time, right?
18:55
The local size and the actual mating size.
18:58
Correct. Under ASME, we have a
19:00
foundational concept often referred to as rule number one
19:03
or the envelope principle. The envelope principle.
19:05
When you specify a size tolerance for a feature of size,
19:08
like a cylinder or a slot, you are simultaneously defining a
19:12
perfect boundary that the part cannot violate.
19:15
OK, so if a pin has a diameter of 10mm + 0 -, 0.1.
19:20
The ASME rule states that the pin can never at any point
19:24
violate a perfect theoretical cylinder of exactly 10.0mm.
19:29
So I take a micrometer which measures A2 point local size and
19:32
I measure 9.9mm everywhere on that pin it passes the local
19:36
size check. Yes, the cross section is good.
19:38
But if that pin was accidentally bent in the machine so it looks
19:41
like a banana, it might not physically fit through a perfect
19:44
10mm hole. Exactly because of rule #1, ASME
19:49
says that banana pin is defective even though every
19:52
single two point micrometer measurement was technically
19:55
within the tolerance band. The perfect form at maximum
19:58
material condition acts as a rigid, unyielding envelope.
20:01
That's it exactly, but here is the trap.
20:05
ISO webinar 1 does not operate with the envelope principle by
20:08
default. Wait really?
20:10
ISO will just let the banana pin pass?
20:12
Yes. In the ISO system, by default a
20:15
size tolerance only controls the two point local size.
20:18
That's. Crazy.
20:19
It only cares what your micrometer says when you pinch
20:21
the part. It does not automatically
20:23
guarantee that perfect outer bounding envelope.
20:25
So how do you ensure the bent pin won't get shipped to you?
20:28
If you want to force the supplier to ensure that the bent
20:31
pin will actually fit into a mating hole, you have to
20:33
explicitly add an additional symbol to the drawing.
20:36
What symbol? An E with a circle around it,
20:39
which stands for the envelope requirement.
20:41
If you as an American engineer used to ASME defaults, send a
20:45
drawing to Europe without that circle E, you are legally
20:48
authorizing them to ship you banana shaped parts as long as
20:52
the cross section measures correctly.
20:54
Wow, this feeds directly into a massive operational hazard, the
20:58
sources highlight, which they call the danger of wrong is
21:01
best. Yes, wrong is best.
21:04
It's such a strange phrase. What does that actually mean in
21:07
the context of a quality control lab?
21:09
It refers to the terrifying illusion of recision created by
21:13
modern coordinate measuring machines, or CMMS.
21:16
Right, those massive robotic arms in the glass rooms.
21:19
Exactly. They are incredibly advanced
21:21
robotic systems. They use a Ruby tipped pro to
21:24
physically tap the part thousands of times, generating a
21:27
digital point cloud with microscopic precision.
21:30
So the machine itself is flawless.
21:31
The machine is Florida State accurate, but the danger lies in
21:35
the invisible software algorithms processing that point
21:38
cloud. Because the software has to be
21:39
told which mathematical dictionary to use.
21:41
Precisely CMM software packages are developed for a global
21:44
market. You might be a machine shop in
21:46
California, inspecting a part explicitly designed to the ASMEY
21:50
14.5 standard, but deep in the settings menu of the CMM
21:55
software. The default evaluation math is
21:58
set to ISO 11 O1. The robot measures the part
22:02
perfectly. The software crunches the
22:04
numbers perfectly, but it uses the ISO algorithm for something
22:08
like position or profile. The screen lights up bright
22:11
green. Big bold letters saying PASS.
22:14
And the machinist celebrates, packs the parts in a crate and
22:17
ships them. Yep.
22:19
But the evaluation was fundamentally mathematically
22:21
wrong according to the legal contract of the drawing.
22:25
It is the ultimate false sense of security.
22:27
You have a half million dollar piece of metrology equipment
22:30
lying straight to your face because of a hidden drop down
22:32
menu. Exactly.
22:33
So what is the hard take away for the listener here?
22:36
How do you defend against this? You have to become absolutely
22:38
militant about the dialect first.
22:40
The specific standard and its revision year must be explicitly
22:44
stated on the drawing or in the models metadata.
22:47
For example, tolerances per ASM. Why 14.5 2018 you?
22:51
Can't just assume they know. Never assume, but that is not
22:55
enough. You must actively audit your
22:58
supply chain and your internal quality labs.
23:01
You have to sit down with the CMM programmers and verify
23:05
visually that their measurement software is explicitly toggled
23:09
to the exact same standard the engineers designed.
23:11
To you have to police the mathematics.
23:13
You cannot assume harmonization, you have to police the math.
23:17
OK, let's assume we have done the hard work.
23:19
We picked a standard, we audited the software, everyone is
23:22
speaking the exact same dialect of GD and T, and there are no
23:26
banana shaped pins sneaking through A.
23:28
Perfect world. Right.
23:29
How do we actually use this language to guarantee that these
23:33
parts assemble smoothly on the factory floor?
23:36
This brings us to the physical reality of fits and the
23:38
fundamental concept of the datum.
23:40
This is where we transition from avoiding disasters to actually
23:44
engineering functionality, right?
23:46
The Zometry sourcing guidelines provide an excellent breakdown
23:48
of this. The entire goal of GD and T in a
23:51
mass production environment is interchangeability.
23:54
Interchangeability. So basically making sure any
23:56
part fits any other part. Exactly.
23:59
It means manufacturing part A in January, manufacturing Part B in
24:03
June, pulling them both blindly out of a warehouse bin in
24:07
December, and knowing with absolute mathematical certainty
24:11
that they will fit together and function identically to every
24:13
other assembly. And the sources make a really
24:16
great point contrasting this with the traditional Craftsman.
24:19
Oh yeah, the rocking horse example.
24:21
Right, if I am a master woodworker building a custom
24:24
rocking horse, I don't need GD and T.
24:28
If I turn a wooden peg on my lathe and it's a little too
24:31
thick to fit into the hole on the rocker, I don't calculate
24:34
the geometric deviation, you just fix it.
24:36
I just grab some sandpaper and manually widen the hole until it
24:39
slides in perfectly. I'm custom fitting every single
24:42
joint. I have the luxury of iterative
24:44
adjustment. But mass production strips that
24:46
luxury away. Completely.
24:48
In mass production you do not have the Craftsman sandpaper.
24:51
You cannot rely on hand fitting because it completely destroys
24:54
the assembly lines throughput. Instead, you rely on calculated,
24:57
standardized fits. And the zometry guidelines
25:00
categorize these into two primary operational modes, right
25:04
clearance fits and interference fits.
25:05
Yes, and the math behind each dictates entirely different
25:09
manufacturing approaches. Let's unpack the clearance fit
25:12
first. Functionally, it is exactly what
25:14
it sounds like. It's an assembly designed so
25:17
that you do not need heavy tools to force the parts together.
25:21
Right, it should assemble freely.
25:23
A clearance fit can range from incredibly loose, like a heavy
25:27
steel bolt dropping easily through an oversized bracket
25:30
hole, to a highly precise lubricated slip fit where you
25:34
slide a ground piston into a cylinder bore by hand and you
25:38
can actually feel the air pressure resisting you because
25:40
of the gap is so microscopic. But the defining characteristic
25:44
is that there was always mathematically a gap.
25:46
Yes. And to guarantee that gap, we
25:49
have to look at the extremes of the manufacturing tolerances,
25:52
specifically a concept called maximum material condition, or
25:56
MMC. Let's pause and define MMC
25:58
clearly because it trips up a lot of people.
26:00
Let's do it. When we say maximum material
26:02
condition, we are talking about the physical state where the
26:05
part weighs the most, right? Right, so for a solid pin, MMC
26:09
is the absolute largest diameter the tolerance allows because a
26:13
fatter pin has more metal. Exactly, but for a hole, MMC is
26:18
the absolute smallest diameter allowed because a smaller hole
26:22
leaves more physical metal in the surrounding block.
26:25
Oh, that's a great way. That is a perfect explanation.
26:28
It is the condition that leaves the maximum amount of material
26:32
on the part. Now, for a clearance fit to be
26:35
mathematically guaranteed, you must design the tolerances so
26:39
that the MMC of the internal part, the absolute fattest
26:42
largest allowed pin, is strictly smaller than the MMC of the
26:46
external part, the absolute tightest smallest allowed hole.
26:49
So if you design the tolerances so that the fattest possible pin
26:52
is still mathematically smaller than the tightest possible hole,
26:55
you have guaranteed a clearance fit no matter where the actual
26:58
manufactured parts fall within their allowed tolerance bands.
27:00
Exactly, it is playing it safe by looking at the worst case
27:03
scenario. Now contrast that with the
27:05
interference fit. This is where the engineering
27:07
gets aggressive. Interference are the exact
27:10
opposite. They're designed to permanently
27:12
lock parts together using immense friction.
27:15
So no gap. No gap.
27:18
To achieve this, the tolerances must be engineered so that any
27:22
physical condition of the internal part is strictly larger
27:26
than any physical condition of the external part.
27:28
So even if I happen to pull the absolutely smallest pin allowed
27:32
by the drawing, and I try to put it into the absolutely largest
27:36
hole allowed by the drawing, the pin is still physically too big
27:40
to fit. They're mathematically designed
27:41
to crash into each other. Exactly.
27:43
They cannot be assembled by hand.
27:44
You have to physically deform the material slightly to get
27:47
them together. Like using an Arbor press.
27:49
Right. You might use a press to
27:51
physically ram a bearing into a housing, or for extreme
27:54
interference fits you use thermal expansion and
27:57
contraction a shrink fit. Oh, I love shrink fits.
27:59
You take the steel pin and submerge it in liquid nitrogen
28:02
to literally shrink its diameter.
28:04
You take the steel housing and hit it with an induction heater
28:07
to expand the hole. And then you work really fast.
28:09
You drop the frozen pin into the hot hole, and as their
28:12
temperatures equalized to ambient room temperature, the
28:15
metal expands and contracts back to its normal size, locking them
28:19
together with thousands of pounds of radial force.
28:22
It's basically welding with temperature and friction,
28:25
basically, yeah. But whether you're designing a
28:27
loose clearance fit or an extreme shrink fit, the geometry
28:31
of these parts is rarely just a simple pin and a simple hole.
28:35
Right, it's never that easy. We're dealing with complex
28:37
gearboxes, engine blocks, and satellite chassis.
28:40
To measure where all these holes and pins are located relative to
28:43
each other, you need a starting point.
28:46
An anchor. Yes, and this is where the
28:48
concept of datum's enters the conversation.
28:51
Our sources emphasize datum's relentlessly.
28:54
Let's demystify them. What exactly is a datum?
28:57
Well, a lot of people try to use the analogy of a house
28:59
foundation. They say a datum is like the
29:01
concrete slab you build from, but a house foundation is
29:05
essentially A2D concept, it only establishes the floor.
29:08
Right, it's just one lane in. Mechanical engineering.
29:11
We operate in 3D space and every part has 6° of freedom.
29:15
It can translate, move up and down, left and right, forward
29:17
and backward, and it can rotate, pitch, roll and Yaw.
29:21
OK, so a datum needs to control. All of that, exactly.
29:24
A datum is a perfect theoretical geometric construct, like a
29:29
plane, an axis, or a center point that we use to arrest
29:32
those degrees of freedom and lock the part in space.
29:35
You know, I've heard a much better analogy for datums.
29:38
Think of pinning A delicate butterfly to a display board.
29:42
Oh, I like where this is going. You have this complex 3
29:44
dimensional object. First you press its back flat
29:47
against the board That is your primary datum plane.
29:49
It takes away 3° of freedom. It can no longer move up and
29:52
down through the board, and it cannot pitch A roll right.
29:55
But it can still slide around on the board, and it can spin like
29:58
a dial. So you push its side against a
30:01
straight edge that is your secondary datum.
30:03
Now can't slide left or right and it can't spin.
30:06
It's getting locked in. It only has 1° of freedom left.
30:09
It can slide up and down along that straight edge.
30:12
Finally, you push the top of the butterfly against a final pin
30:15
that is your tertiary datum. The butterfly is now fully
30:18
locked in 3D space. That is an incredibly accurate
30:21
mechanical analogy. What you just described is the
30:25
fundamental 321 rule of datum reference frames. 321.
30:29
By establishing a primary, secondary, and tertiary datum,
30:32
you lock the part into a theoretical corner.
30:35
Once the part is completely immobilized in that perfect
30:38
theoretical reference frame, then and only then can you start
30:41
measuring the true position position of every other whole
30:43
slot and surface on the part. Because now everyone agrees on
30:46
exactly where 0 is. Exactly.
30:48
The data reference frame serves as the universal anchor for the
30:51
designer, the machinist, and the quality inspector.
30:53
It completely eliminates the tribal knowledge ambiguity of
30:56
where to measure from. So on the factory floor, a
30:59
theoretical datum plane isn't just an idea, it is simulated by
31:02
physical equipment, right? You take your physical, slightly
31:06
warped imperfect machine block and you rest it on a massive,
31:11
perfectly calibrated surface plate.
31:13
That physical contact between the imperfect part and the
31:16
perfect granite establishes datum A.
31:19
Yes. Now every height gauge merriment
31:22
taken from that granite clate upwards is anchored to the exact
31:25
same universal mathematical truth.
31:27
Exactly, the drawing tells the inspector.
31:29
Explicitly. Press this specific face against
31:32
the granite. Push this specific edge against
31:34
the angle block. Now measure the hole.
31:36
OK, this raises a massive question for me and I really
31:39
want to push back on the psychology of design engineers
31:41
here. Because if datums are this
31:43
perfect universal anchor and they provide absolute
31:46
mathematical clarity, why do we still see so many parts costing
31:50
a fortune Because the tolerances are insanely unnecessarily
31:54
tight. Is it possible that most over
31:57
tolerancing where an engineer demands a tolerance of 0.0001
32:01
inches on a non critical part is actually just a symptom of them
32:05
fundamentally misunderstanding datums?
32:07
You have hit on one of the most common and expensive
32:10
psychological failures in design engineering.
32:13
Yeah, yes, chronic overtolerancing is very
32:16
frequently a symptom of datum panic.
32:18
Datum panic. I love that term.
32:20
How does it happen? Imagine an engineer is designing
32:23
a long cantilevered bracket. They choose a very small,
32:26
somewhat rough surface. At the very back of the bracket
32:29
is their primary datum. OK, a bad foundation.
32:32
A terrible foundation because that starting face is small and
32:36
imperfect. It doesn't stabilize the part
32:38
well. Any microscoic variation at that
32:40
small datum face gets mathematically amplified the
32:43
further out you measure along the bracket.
32:45
It acts like a lever arm of error.
32:47
Oh, so a tiny tilt at the base means the tip is way off?
32:51
Exactly. The engineer runs a basic
32:53
analysis in their CAD software, sees this massive amplified
32:56
variation of the far end of the bracket, and realizes the mating
32:59
holes aren't going to align. And instead of stepping back and
33:02
saying wait, I chose a terrible foundation, let me pick a better
33:04
datum structure. They panic.
33:07
They panic. They leave the bad datum in
33:09
place and their reflex is to just crush the linear
33:12
tolerances. They demand that the machine
33:15
shop hit impossible levels of precision to compensate for the
33:19
shaky foundation. So they demand aerospace level
33:22
precision on a bracket that just holds a sensor, totally
33:26
skyrocketing the manufacturing cost.
33:28
Precisely as the sources strongly warn, we must avoid
33:32
over tolerancing a functionally logical datum reference frame,
33:36
one that mimics exactly how the part will physically bolt into
33:39
the final assembly, prevents the need for impossible precision.
33:43
Because if you anchor the part correctly based on its actual
33:46
physical function, you can absorb a lot of natural
33:49
manufacturing variation. You can absorb a ton of it.
33:51
You can loosen the tolerances, save a massive amount of money
33:54
on machining and inspection, and still mathematically guarantee
33:58
interchangeability. Fix the foundation and you can
34:00
relax the walls. That is a profound paradigm
34:03
shift. Tolerances are tight because the
34:05
foundation is shaky, but you know anchoring a single part to
34:08
a granite table only solves half the problem.
34:10
What happens when you take 510 or 50 of those perfectly datum
34:14
parts and try to bolt them all together in a single assembly?
34:17
The errors accumulate. Exactly.
34:20
And this brings us to the most technically dense,
34:22
mathematically rigorous part of our deep dive today, the Boeing
34:26
Research and Technology paper by Fritz Schultz on tolerance stack
34:30
analysis. This is where GD and T
34:33
transitions from pure 3D geometry into the realm of
34:36
statistical mathematics. Yes, the heavy math.
34:39
When you are assembling a complex system, you are not just
34:42
verifying individual parts, you are evaluating an assembly gap.
34:46
We want to know if part A+, Part B plus Part C bolted together
34:50
will still fit inside the main housing or if the accumulated
34:53
variation will cause them to crash.
34:55
And in Schultz's paper, the critical gap we are trying to
34:58
evaluate is referred to mathematically as G, right?
35:00
And this gap G is usually just a function, often a simple sum or
35:04
difference of the individual detailed dimensions of all the
35:07
parts in the stack, which Schultz denotes mathematically
35:10
as Li. So the total length of the stack
35:12
is L1 plus L2 plus L3 and so on. Correct.
35:16
And the entire discipline of tolerance stack analysis is
35:19
about predicting what happens to that final gap G when every
35:23
single one of those Li dimensions inevitably varies
35:27
within their allowed tolerance zones.
35:28
Because they will vary. Always.
35:31
Schultz outlines 2 primary methods for calculating this,
35:34
and they represent completely different philosophies of risk.
35:37
Let's start with method one worst case analysis, which is
35:40
mathematically referred to as arithmetic tolerancing all.
35:43
Right, Walk me through the worst case scenario.
35:45
Just how pessimistic is this math?
35:48
It is the absolute maximum level of pessimism in arithmetic worst
35:52
case analysis. The engineer assumes that every
35:55
single part in the assembly has been machined to the extreme
35:58
absolute limit of its tolerance band.
36:00
Ouch. Furthermore, it assumes that
36:02
nature is actively conspiring against you, and every single
36:05
part happens to be at the extreme limit in the exact same
36:08
direction. So if I'm stacking 5 steel
36:10
plates and the drawing says each plate can be up to 0.1
36:13
millimeters too thick, worst case analysis assumes that the
36:16
machine is didn't hit the nominal target on a single
36:18
plate, not one. It assumes plate 1 is exactly
36:21
0.1 millimeters too thick, plate 2 is exactly 0.1 millimeters too
36:27
thick, and so on. Exactly.
36:29
You simply Add all the individual extreme tolerances
36:32
together using basic arithmetic to find the total accumulated
36:35
maximum variation which salts denotes mathematically as delta
36:40
Y. OK, so delta Y.
36:42
In your example, Delta Y is 0.5mm.
36:46
You then design your main housing to accommodate that
36:48
massive 0.5mm worst case growth. But logically, that seems
36:53
completely unsustainable from a cost perspective.
36:55
That is, if I'm a project manager and my engineers are
36:58
designing based on arithmetic worst case scenarios, I am
37:01
paying a fortune. The statistical odds of reaching
37:03
your hand into five different bins of parts pulled from
37:06
different manufacturing batches and randomly selecting 5 parts
37:10
that are all simultaneously maxed out at the extreme upper
37:13
limit of their tolerance. The odds are infinitesimally
37:16
small. Right.
37:17
If I force the machine shop to hold insanely tight tolerances
37:21
on the individual parts just to make this worst case sum look
37:24
manageable on paper, I am bleeding the budget dry buying
37:28
precision I don't actually need. You have perfectly articulated
37:31
the inherent flaw of arithmetic tolerancing in mass production.
37:35
It is incredibly safe, but it is ruinously expensive.
37:39
That economic reality drove the industry to adopt Method 2, the
37:43
RSS method, which stands for root sum square or statistical
37:48
tolerance. OK, so RSS brings the math back
37:51
to reality. How does it work?
37:53
RSS acknowledges that manufacturing processes, when
37:56
operating correctly, follow a normal distribution, a bell
37:59
curve. Right, most parts hit near the
38:01
middle. Exactly.
38:02
Most of the parts the CNC machine produces will fall right
38:04
near the perfect nominal center of the tolerance band.
38:07
A few will be a little larger, a few will be a little smaller,
38:10
but the extreme limits are rare. So it's about probability, yes.
38:13
RSS uses the standard deviation, denoted mathematically as Sigma,
38:17
and the statistical variance of the manufacturing process to
38:20
predict the stack. So instead of assuming
38:22
everything goes perfectly wrong in the same direction, RSS
38:25
assumes A cancelling effect. Exactly.
38:28
If part A happens to be a little too thick, statistical
38:31
probability says Part B is likely near the middle, and part
38:35
C might actually be a little too thin.
38:38
They naturally absorb each other's variations in the final
38:40
assembly. Exactly, by taking the square
38:43
root of the sum of the squares of the individual standard
38:46
deviations, RSS provides a much tighter, more realistic
38:50
prediction of the final assembly variation.
38:53
Which means cheaper parts. Yes, because the predicted
38:57
assembly tolerance is tighter, the engineer can safely loosen
39:01
the tolerances on the individual detail parts.
39:04
Looser parts are cheaper to machine, cheaper to inspect, and
39:07
you save massive amounts of money without sacrificing the
39:09
assembly. It sounds like the perfect
39:11
solution, but the Boeing paper doesn't stop at basic RSS.
39:15
Schulz throws a massive wrench into this clean mathematical
39:18
bell curve. Oh, the mean shifts.
39:20
Yes, he introduces the reality of mean shifts.
39:23
And this is where I need you to clarify something, because
39:25
reading the paper I got a bit confused.
39:26
Sure, tool wear is a physical measurable thing, right?
39:30
A carbide cutting insert loses a Micron of its edge every time it
39:34
passes over the steel. That is a linear, predictable
39:38
physical change. The tool gets shorter, so the
39:41
parts slowly get larger, right? Why on earth is Schultz treating
39:45
a predictable physical degradation as a statistical
39:47
variable that ruins the RSS math?
39:50
That is an excellent push back and it gets to the heart of why
39:53
the Boeing paper is so important.
39:55
You are right, the wear on one single tool cutting one single
39:58
part is linear and predictable. OK, but remember we are talking
40:02
about mass production aerospace supply chains.
40:05
When Boeing orders 10,000 specialized fasteners, they
40:08
aren't coming off 1 machine. Right, they're outsourced.
40:11
They're coming from 5 different CNC lathes operating in three
40:14
different supplier facilities, running at slightly different
40:16
feed rates, operated by different machinists who change
40:19
out the dole cutting tools at different intervals.
40:22
I see it's a chaotic system. So when you aggregate all the
40:25
those parts into a single warehouse bin, the overall
40:28
distribution of those parts it's still a bell curve, but the
40:32
center of that bell curve, the mean, is no longer perfectly
40:36
centered on the nominal dimension.
40:38
Because tools wear down before they are replaced.
40:40
Exactly because of that, the average size of the parts tends
40:44
to drift toward the side of the tolerance band that represents
40:47
less material removal. That is a mean shift.
40:50
The bell curve is physically sliding to the right or the left
40:53
overtime. So you can't rely on pure basic
40:56
RSS statistics, because the fundamental assumption of RSS
40:59
that the variations will randomly cancel each other out
41:01
around a perfect center point is physically false.
41:04
The deck is stacked in the direction of the tool wear.
41:07
Precisely. And this is where Schultz
41:08
details the truly fascinating math of risk assessment in
41:12
traditional centered statistical analysis.
41:14
Engineer's design for a ± 3 Sigma range.
41:17
OK ±3 Sigma. In a perfectly centered bell
41:21
curve, this corresponds to an accepted failure rate of 0.27%,
41:27
meaning they accept the risk that roughly 3 out of every 1000
41:31
assemblies will statistically fall outside the limits and fail
41:34
to assemble. And they accept that tiny risk
41:37
because it saves millions in manufacturing costs compared to
41:40
worst case tolerance saying exactly.
41:42
But wait, if the mean shifts if the whole bell curve slides to
41:46
the right because the tools are wearing down, doesn't that
41:49
massively increase the failure rate on the right side of the
41:52
curve? It does push the curve into the
41:54
upper limit, but here is the brilliant counterintuitive
41:58
mathematical revelation in the balling paper.
42:00
OK, lay it on me. When you take the worst case
42:03
mean shifts, assuming the tool wear on all the parts drifted in
42:06
the absolute worst possible direction, and you stack those
42:09
arithmetic shifts on top of the RSS statistical variability, the
42:14
overall risk of random failure doesn't skyrocket, it actually
42:17
drops. Wait, how is that mathematically
42:19
possible? You're adding worst case to RSS.
42:21
Because the risk is no longer spread across both extremes.
42:25
In a perfectly centered bell curve, you have a tiny risk of
42:28
failing by being too small and a tiny risk of failing by being
42:32
too big. But when the mean shifts
42:35
severely to the too big side, the tail of the bell curve on
42:39
the too small side gets pulled completely away from the lower
42:42
limit. You essentially eliminate the
42:44
risk of the assembly being too small.
42:46
All of your risk is now concentrated solely on the too
42:49
big upper limit. That is wild.
42:51
Because you were only fighting one tail of the distribution
42:54
instead of 2, the total overall probability of failure
42:57
essentially halves. The risk factor drops from 0.27%
43:02
down to 0.135%. The predictable physical reality
43:06
of the tool wear actually eliminates one of the
43:08
statistical failure modes, having your overall random risk
43:11
to 0.135%. Yeah, it is brilliant.
43:15
It is brilliant. So what does Boeing do with this
43:17
information? Do they just enjoy a slightly
43:19
lower failure rate on the assembly floor?
43:21
No, an organization like Boeing uses that mathematical risk
43:25
reduction as leverage. They use it to buy back even
43:28
more tolerance for the machine shops to save even more money.
43:31
So they tweak the math. Schultz explains that to
43:33
seamlessly take advantage of this risk reduction while
43:36
maintaining the traditional 0.27% failure target, they
43:40
inject a specific mathematical correction factor into the RSS
43:44
equation. What's the factor?
43:46
The RSS portion of the part to part variation is multiplied by
43:49
an exact factor 0.927. 0.927 Yes.
43:54
By multiplying the RSS variation by 0.927, you artificially
43:59
shrink the statistical variance prediction just enough to
44:02
mathematically balance out the worst case mean shifts.
44:05
It allows the engineer to account for the gritty reality
44:08
of tool wear while safely maximizing the allowed
44:10
tolerances, all without exceeding the accepted failure
44:13
rate. It is an absolute master class
44:15
in blending the pessimism of worst case arithmetic with the
44:19
optimism of pure statistics. But let's bring this down to the
44:22
desk of the everyday engineer. If I'm sitting in front of my
44:26
CAD terminal trying to design A gearbox assembly, how do I
44:30
practically apply this? Good question.
44:32
Do I run the ultra pessimistic worst case Delta Y calculation
44:37
and overpay for parts? Or do I run this highly complex
44:41
Boeing RSS math with the 0.927 multiplier and risk that a mean
44:46
shift might occasionally cause an assembly line jam?
44:49
Well, the Boeing paper actually provides a highly pragmatic,
44:52
straightforward compromise for the working engineer.
44:54
Oh, thank goodness. Scholl suggests that for
44:56
critical gaps, you should actually calculate the tolerance
44:58
stack using both methods. Both of them.
45:01
Yes, you run the arithmetic worst case stack to see the
45:05
absolute physical limits. Then you run the modified RSS
45:08
stack with the mean shifts to see the statistical probability.
45:11
OK. Finally, you take those two
45:13
resulting gap numbers and you simply average them together.
45:15
It is brilliant in its simplicity.
45:17
You just find the exact midpoint between absolute paranoia and
45:22
absolute mathematical optimism. Exactly.
45:25
Averaging the two provides a highly robust balance tolerance
45:29
limit that protects your assembly from catastrophic worst
45:33
case stack UPS while explicitly preventing you from forcing the
45:37
machine shop to achieve impossible budget destroying
45:40
precision. That right there is the kind of
45:43
practical engineering philosophy that saves companies millions of
45:46
dollars. Calculate both extremes,
45:48
understand the physics behind both, and engineer the middle
45:51
ground. Now we are entering the final
45:53
section of our deep dive today, and we want to deliver exactly
45:57
on the mission we set out at the beginning.
45:59
We're going to synthesize all this heavy mathematics and
46:02
historical context to answer the most common burning questions
46:06
people in the industry are asking right now.
46:07
It is the perfect way to wrap this up.
46:09
We've covered the theory, but how does this apply to the
46:11
rapidly evolving future of manufacturing technology?
46:15
Let's start with question one, which is a massive debate driven
46:18
by software vendors. A lot of tech companies are
46:20
pushing this concept of the digital twin, right?
46:23
Yes. Digital Twin.
46:25
So will advanced 3D CAD, solid models and AI eventually
46:30
eliminate the need for GD and T altogether?
46:33
Will the software just handle all the variation in the
46:35
background without us having to define datum's and tolerance
46:38
zones? The answer from the sources is
46:40
an absolute emphatic no. Not going to happen.
46:43
The ASQ overview document explicitly addresses this
46:47
software driven utopian vision. It acknowledges that future
46:51
technology embedded directly into SOLID models will
46:54
absolutely change how we apply and process the rules.
46:57
Like automating the math. Right, the software will enable
47:00
fully automated tolerance stack analysis with a click of a
47:02
button. It will allow for highly
47:04
advanced FEA finite element analysis.
47:07
And for those who might not know, FEA is where the computer
47:09
breaks a complex 3D shape down into thousands of tiny digital
47:13
mesh blocks and mathematically simulates exactly how the part
47:16
will bend, stretch and distribute stress when physical
47:19
loads and heat are applied to. Exactly, and future software
47:22
will be able to run that FEA simulation not just on the
47:25
perfect nominal part, but by actually simulating the maximum
47:28
feature variation allowed by the tolerances.
47:30
Which is amazing. It is.
47:31
It will push the industry into a truly paperless, model based
47:35
definition environment. The physical 2D blueprint will
47:38
die. But the concept of tolerance
47:41
less design, the idea that the software will just figure it out
47:43
on its own, is a massive, dangerous misconception.
47:47
The source is incredibly blunt about this.
47:49
It states, and I quote tolerance less design.
47:52
No way. Complete with multiple
47:54
exclamation. Points and you cannot code away
47:56
physics. Exactly.
47:57
The 3D digital model is perfect. It is dimensionless math and
48:01
avoid. But the physical CNC machine
48:04
cutting the steel is not perfect.
48:06
The tool will still wear down. The titanium will still warp
48:09
when it gets hot during machining, the spindle will
48:12
still vibrate. The rules of GD and T, the datum
48:15
reference frames that establish the physical anchor, the
48:17
geometric tolerance zones that define acceptable limits of
48:20
physical form. Those rules must fundamentally
48:23
exist. Even if they're invisible.
48:25
Yes, they will exist as invisible metadata attached to
48:29
the faces of the 3D solid model, rather than symbols printed on a
48:32
piece of paper, but the engineer must still actively define the
48:36
functional limits of the hardware.
48:38
The software will automate the math, but it cannot invent the
48:42
design intent. Which perfectly sets up question
48:44
too, who actually needs to use GD and T in the modern era?
48:49
Is this strictly an aerospace thing for people building
48:51
rockets at 10,000 a pound? Or does the engineer designing a
48:55
mountain bike suspension hub need to use this complex
48:58
language too? It brings us entirely back to
49:00
that core philosophy of interchangeability.
49:02
Right, the rocking horse versus mass production.
49:04
The sources are clear. If you are a Craftsman hand
49:07
fitting a custom assembly, or if your entire product is machined
49:11
by one guy on one lathe who checks the fit of every part
49:15
manually, you do not need GD and T.
49:17
Save yourself the headache. Exactly.
49:20
But if you are manufacturing at any kind of scale, it is
49:22
mandatory, whether you are building space station
49:25
components, mountain bike hubs or consumer blenders.
49:27
Because of the supply chain, yes.
49:29
If you rely on a decentralized supply chain where the housing
49:33
is cast in Mexico, the gears are machined in Germany, the
49:35
bearings are bought off the shelf from Japan, and it is all
49:38
bolted together in an assembly plant in Texas.
49:41
GD and T is the only legal mathematical language that
49:45
guarantees those parts will fit. Without tribal knowledge.
49:48
Exactly. It is the only way to
49:50
mathematically guarantee interchangeability across
49:53
language barriers, time zones and different manufacturing
49:56
processes. And that leads us to the
49:58
ultimate question. Question 3.
50:00
How do we fix the Houston it doesn't fit problem today?
50:05
Actionable advice, right? If a project manager listening
50:08
to this right now is struggling with massive scrap rates, parts
50:11
that won't assemble, and vendors arguing over inspection reports,
50:15
what is the immediate actionable prescription based on these
50:19
sources? The prescription requires a hard
50:21
reset of the engineering culture.
50:23
First, you must immediately and systematically eradicate tribal
50:26
knowledge. Write it all down.
50:27
Yes. If your quality inspector or
50:29
your machinist has to walk over to an engineer's desk to ask
50:32
what a specific dimension actually means, your
50:35
documentation system is fundamentally broken.
50:38
Agreed. Second, you must adopt and
50:40
ruthlessly enforce an internationally recognized
50:43
geometric standard. The sources highlight ASMEY
50:46
14.52 out of 18 as the gold standard for robust functional
50:51
design. And as we learn with the 65%
50:53
problem, you can't just assume everyone knows you're using it.
50:56
Absolutely not. Step 3 is explicit verification.
51:00
You must explicitly state which standard, including the specific
51:03
year of revision, governs every single data set, CAD model, and
51:07
drawing you send out. Right Tolerances per ASMEY 14.5,
51:11
2018. And you must actively verify
51:13
that your suppliers CMM software is set to calculate using that
51:16
exact same dialect. Do not let wrong as best hide in
51:19
the software defaults. Audit the software.
51:21
Audit everything. And finally you have to elevate
51:24
your engineering departments mathematical rigor.
51:27
They must understand whether a specific critical interface
51:30
requires A rigid worstcase arithmetic stack analysis, or if
51:34
the company can save massive amounts of capital by utilizing
51:38
statistical RSS methods with the proper Boeing mean shift
51:42
multipliers. It is about moving from an
51:45
attitude of hoping the metal fits to an attitude of
51:48
mathematically proving it will fit before the purchase order is
51:51
even signed. Precisely, it is about total
51:54
intentionality. Which brings me to a final
51:56
thought. I want to leave you with
51:57
something to genuinely Mull over the next time you're staring at
52:00
a complex 3D model on your screen or just walking the floor
52:03
of machine shop. Ultimately, GD and T is not
52:06
about demanding absolute perfection from the tools and
52:09
the humans that make our parts. In fact, it's the exact opposite
52:12
of that it is. It's an incredibly brilliant,
52:15
globally evolved mathematical framework that exists solely to
52:19
allow for human imperfection. It's a system that accepts that
52:22
tools will dull, metal will warp, and machines will vibrate.
52:26
And yet it still guarantees that the final complex machine will
52:30
bolt together and function beautifully.
52:32
That's a great way to frame it. It makes you wonder is the true
52:35
highest art of engineering not the pursuit of perfection, but
52:39
rather the absolute mastery of acceptable flaws?
52:43
That is a profound perspective. It is the rigorous science of
52:46
knowing exactly how imperfect you can afford to be and
52:50
building a world that works perfectly within those
52:52
boundaries. If you got something useful out
52:53
of today's episode, hit follow and drop a like.
52:55
It helps grow the show and keeps the engineering knowledge
52:58
flowing.